JP7452668B2 - Task management device, task management method, and task management program - Google Patents

Description

The present invention relates to a task management device, a task management method, and a task management program.

With the development and spread of technologies such as IoT (Internet of Things), big data utilization, and autonomous driving, power consumption at data centers accounts for approximately 1% of the world's power consumption, and is expected to increase further in the future. Ru. A data center has a mixture of servers equipped with different CPUs, such as Intel (registered trademark) and ARM (Acorn RISC Machine) (registered trademark).
Therefore, Non-Patent Documents 1 and 2 describe methods for satisfying performance requirements while suppressing power consumption as scheduling techniques for deploying (arranging) VMs (Virtual Machines) and containers on each server.

Rocha, Isabelly, et al. "HEATS: Heterogeneity-and Energy-Aware Task-based Scheduling." 2019 27th Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP). IEEE, 2019. Ismaeel, Salam, Ali Miri, and Ayman Al-Khazraji. "A novel host readiness factor for energy-efficient VM consolidation in cloud data centers." 2019 8th International Conference on Modeling Simulation and Applied Optimization (ICMSAO). IEEE, 2019.

Pollution of the global environment due to carbon dioxide emissions during power generation is considered a worldwide problem. Therefore, there is a shift from conventional thermal power generation using fossil fuels to power generation using renewable energies such as solar power generation and wind power generation, which have less impact on the environment.
Here, since renewable energy depends on the natural environment such as season and region, the amount of power generation is unstable. Therefore, there is a need for a scheduling technology that takes into account the unstable power supply of renewable energy.

However, conventional scheduling techniques such as those disclosed in Non-Patent Documents 1 and 2 are based on stable power supply, and cannot cope with unstable power supply systems such as temporary evacuation of tasks during power shortages.

Therefore, the main objective of the present invention is to appropriately allocate tasks even in an unstable power supply source.

In order to solve the above problems, a task management device of the present invention has the following features.
The present invention provides a power generation amount collection unit that calculates the usable power amount of a server group based on the power amount obtained by subtracting the amount of power used to be stored in a battery from the power generation amount of a power supply unit that is a regenerative power generation plant;
a usage collection unit that calculates the power usage of the server group;
Even if the stored power of the battery is used as a backup power source, if the available power amount is insufficient for the amount of power used, the currently running task is interrupted according to the amount of power shortage, and the interrupted task is and a task control unit that stops the server that was running the server and migrates the task on the server that is currently being stopped to a more power efficient server that is searched from a set of servers that will not be stopped this time. .

According to the present invention, tasks can be appropriately allocated even in an unstable power supply source.

FIG. 1 is a configuration diagram of a task processing system according to the present embodiment. FIG. 2 is a hardware configuration diagram of a controller according to the present embodiment. 7 is a flowchart showing processing of a controller according to the present embodiment. It is a graph showing the relationship between the amount of available power and the amount of power used according to the present embodiment. 3 is a table showing an example of tasks initially arranged in a server group according to the present embodiment. It is a table showing what to do when the power shortage is 100 [Wh] in classification 3 according to the present embodiment. It is a table showing what to do when the power shortage is 150 [Wh] in classification 3 according to the present embodiment. It is a table showing what to do when the power shortage is 200 [Wh] in classification 3 according to the present embodiment. 9 is a table showing what to do when the available power recovers from the state shown in FIG. 8 according to the present embodiment, resulting in a classification 1 power surplus state.

Hereinafter, one embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a configuration diagram of a task processing system 100.
The task processing system 100 includes a data center 1 and a regenerative power generation plant (power supply unit) 3 that supplies power to the data center 1. The renewable power generation plant 3 is a facility or equipment that generates power using renewable energy such as solar power generation and wind power generation.

The data center 1 includes a server group 21 that processes tasks, a battery 22 that supplies power to the server group 21 in the event of a power outage in the regenerative power generation plant 3, and a controller (task management) that assigns each task to each server constituting the server group 21. Apparatus) 10. A task is a processing unit of software that operates to provide a service. For example, a VM (Virtual Machine) or container running on a server is an example of a task.
The server group 21 is a set of servers to which each task is assigned, and each server may be a different type of server than other servers. A heterogeneous server is one in which hardware resources such as CPU, memory, storage, etc., which are the components of the device, are different from other servers.

The controller 10 includes a placement calculation section 11 , a task storage section 12 , a usage amount collection section 13 , a task control section 14 , and a power generation amount collection section 15 .
The usage collection unit 13 collects the resource (power) usage of each operating server in the server group 21 as statistical information. The usage collection unit 13 also calculates the free resource capacity of each server by subtracting the resource usage from the total resource amount of each server.
The placement calculation unit 11 determines which task should be deployed (reserved) on which server of the server group 21 at the time of service start to achieve optimal placement, based on the resource free capacity of each server determined by the usage amount collection unit 13. to schedule.
The task storage unit 12 stores data on tasks to be placed by the placement calculation unit 11.

The power generation amount collection unit 15 constantly monitors the power generation amount of the regenerative power generation plant 3 and collects the results as statistical information.
The task control unit 14 determines the amount of power available for processing the task from the collected amount of power usage and amount of power generation. Then, the task control unit 14 appropriately changes the arrangement of tasks and the operating status of the server according to the amount of available power (or changes in the amount of power generation).
For example, when the amount of available power decreases, the task control unit 14 stops a low-priority task and the server that was running the task in accordance with the power shortage. On the other hand, when the amount of available power increases, the task control unit 14 causes tasks with low priority to operate on the server in addition to tasks with high priority.

FIG. 2 is a hardware configuration diagram of the controller 10.
The controller 10 is configured as a computer 900 having a CPU 901, a RAM 902, a ROM 903, an HDD 904, a communication I/F 905, an input/output I/F 906, and a media I/F 907.
Communication I/F 905 is connected to external communication device 915. The input/output I/F 906 is connected to the input/output device 916. The media I/F 907 reads and writes data from the recording medium 917. Further, the CPU 901 controls each processing unit by executing a program (also called an application or an abbreviation thereof) read into the RAM 902 . This program can also be distributed via a communication line or recorded on a recording medium 917 such as a CD-ROM.

FIG. 3 is a flowchart showing the processing of the controller 10.
The task control unit 14 reserves a task to be operated on each server of the server group 21 of the data center 1 (S101). This allows unplaced tasks (reserved tasks) in the task storage unit 12 to be deployed at the reserved time (service start time).
When the current time reaches the reservation time, the usage collection unit 13 checks the available resource capacity of each server (S102). The task control unit 14 selects the confirmed server with sufficient resource free capacity as a candidate server to which the task will be added.

The placement calculation unit 11 calculates the placement of reserved tasks suitable for the available resource capacity of the candidate server (S103). In the calculation at S103, the placement calculation unit 11 may apply an optimal placement search technique using an approach using linear regression, a neural network, or the like.
The task control unit 14 deploys the reserved task to the destination server based on the calculation result of S103 (S104).
As described above, the initial arrangement of tasks has been completed through the processes of S101 to S104. However, since the power generation amount of the regenerative power generation plant 3 changes from time to time, a process for optimizing the arrangement of tasks according to the power generation amount monitored by the power generation amount collection unit 15 will be described from S110 onwards.

The task control unit 14 determines the amount of power that can be used by the server group 21 for task processing (hereinafter referred to as “usable power amount”) based on the power generation amount of the renewable power generation plant 3 collected by the power generation amount collection unit 15, for example, using the following formula. ``Electric energy'').
(Amount of available power) = (Amount of power generated by the renewable power generation plant 3) - (Amount of power used by equipment in the data center 1 other than the server group 21, such as the controller 10) - (Amount of power used to be stored in the battery 22)

Next, the task control unit 14 classifies the current power supply status as follows by comparing the calculated available power amount with the power usage amount of the server group 21 collected by the usage amount collection unit 13. (S110).
(Category 1) If available power≫power consumption, it is determined that there is a “surplus power” and the process proceeds to S121.
(Category 2) If the amount of available power≈the amount of power used, it is determined that the balance is good, and the process proceeds to S131.
(Category 3) If available power << power usage, it is determined that there is a "power shortage" and the process proceeds to S141.
Note that the power consumption of the server group 21 is a statistical value of the power amount required when each server is operating (the sum of the power amount of all servers), and may be measured as an average value, mode, or minimum value, for example. be done. Further, in the case of the usable power amount << the power usage amount (=minimum value) in category 3, the power shortage is especially serious, so it is desirable to use the stored power of the battery 22 as a backup power source.

FIG. 4 is a graph showing the relationship between available power amount and power usage amount. The horizontal axis is the amount of available power, and the value increases as you move to the right. Further, the average value of the power consumption of the server group 21 is assumed to be V2 on the graph.
If the available power amount is equal to or higher than V3, it greatly exceeds V2, so the task control unit 14 determines it to be a category 1 "power surplus."
If the available power amount is V1 or more and less than V3, there is little excess or deficiency from V2, so the task control unit 14 determines that it is classified as "well-balanced" in category 2.
If the available power amount is less than V1, it is significantly less than V2, so the task control unit 14 determines it to be a category 3 "power shortage."
Here, the threshold value (V3-V2) and the threshold value (V2-V1) based on V2 are respectively calculated from the variance of the average power usage of the server group 21, and the threshold value (V2-V1)>threshold value (V3 -V2).

Hereinafter, returning to FIG. 3, a process in which the task control unit 14 appropriately changes the arrangement status of tasks and the operating status of servers according to each classification will be specifically described.
(Category 3) Processing for “power shortage” will be explained.
The task control unit 14 suspends the service of the task according to the amount of power shortage (=power usage amount - available power amount), creates a snapshot of the current state of the suspended task, and temporarily saves the current state of the suspended task in the task storage unit 12. Store it (S141). As a result, even if the server that was running the interrupted task is stopped, the current state of the task is not lost.

Then, the task control unit 14 reduces the amount of power used and improves the "power shortage" state by stopping the server for which there are no running tasks through the task interruption process in S141 (S142). The process of stopping the server in S142 is not limited to the process of turning off the power, but may be a process of reducing power usage such as transitioning to a sleep state. Then, the task control unit 14 returns the process to S110.
Here, the task control unit 14 will clarify which task is to be interrupted in S141 and which server is to be stopped in S142 in the specific example shown in FIG. 5 and thereafter.

(Category 1) Processing of "surplus power" will be explained.
The task control unit 14 restarts the stopped server by executing the process of S142 (S121). This allows the task to be deployed on the restarted server. Then, the task control unit 14 reads from the task storage unit 12 the task that has been suspended due to the execution of the process in S141, and reschedules the task to run on the server restarted in S121 (S122). .
When the rescheduled execution time comes, the rescheduled task is read from the task storage unit 12 and deployed on the server. As a result, the deployed task is deleted from the task storage unit 12. Then, the task control unit 14 returns the process to S102.

(Category 2) “Good balance” processing will be explained. Note that when the balance is good, there is little excess or deficiency in the amount of electric power, so there is no need to urgently change the arrangement of tasks or the operational status of the server. Therefore, the task control unit 14 may omit the task placement optimization processing described in S131 to S133 below.

The placement calculation unit 11 calculates a task placement suitable for the available resource capacity of the candidate server (S131). Note that the process of searching for the optimal placement of tasks is common in S103 and S131. On the other hand, in S103, a destination server is determined for a reserved task that has not yet been placed, but in S131, a task that has been placed is set as a migration task to be migrated (moved), and a destination server is determined.
The task control unit 14 interrupts the migration task running on the migration source server and saves it to the task storage unit 12 (S132). Then, the task control unit 14 reads the moving task being evacuated from the task storage unit 12, and re-reserves the moving task to operate on the destination server (S133). Then, the task control unit 14 returns the process to S104.

Hereinafter, the arrangement status of tasks and the operating status of the server will be specifically explained with reference to FIGS. 5 to 9.
FIG. 5 is a table showing an example of tasks initially placed in the server group 21. As shown in FIG. This table associates, for each server, the currently placed task, the priority of the task, the priority of the server according to the priority of the task, and the power consumption of the server.
Note that the power that can be collected as power consumption of the server is the sum of power consumption during idle time and power consumption associated with processing of each task. In addition, for both task priority and server priority, the larger the numerical value, the higher the priority, and the harder it is for important tasks or servers to be stopped even in the event of a power shortage.

For example, in the server P1, a task T1 with priority=7 (highest) is running, and the power consumption of the entire server is 200 [Wh]. On the other hand, in the server P3, task T2 with priority=4 and task T3 with priority=6 are operating, and the power consumption of the entire server is 300 [Wh]. Furthermore, although server P2 is not running any tasks, the server itself is powered on and its power consumption is 100 [Wh].
In other words, the power consumption of servers P1 to P3 is high (more than 100), so although their processing capacity is high, their power efficiency is poor. The power consumption of servers P4 to P6 is low (less than 50), so although their processing capacity is low, their power efficiency is good.

First, the task control unit 14 calculates the priority of tasks running on each server. For example, one or a combination of the following can be used as input parameters for calculating priorities:
- Input parameter indicating the usage rate of the task (service). For example, the higher the server's CPU usage rate or the server's memory usage rate, the higher the task priority.
- Input parameter indicating the degree of urgency inferred from the requirements for the task (service). For example, the more severe the required response time and real-time processing performance, the higher the task priority.

Then, the task control unit 14 obtains the highest priority value of the tasks running on each server, and assigns higher priorities to the servers in descending order of the highest value. For example, the highest task priority for server P3 is MAX(4,6)=6, so server priority = 5 is given next to server P1 (task priority = 7), which has server priority = 6. Ru.

FIG. 6 is a table showing what to do when the power shortage is 100 [Wh] in classification 3.
The table in FIG. 6 has the same format as the table in FIG. 5, with an explanatory "change" column added to the rightmost column. This change column shows the contents of changes from the state of FIG. 5.
The task control unit 14 adds the power consumption of the servers in descending order of server priority until the power consumption corresponds to the power shortage. In the example of FIG. 5, the power consumption of the server P2 with the lowest server priority=1 is 100, so the task control unit 14 only needs to stop the server P2.
The task control unit 14 temporarily stores the task on the server P2 to be stopped in the task storage unit 12 (S141), but since there is no task on the server P2 this time, the process of S141 is omitted. Then, the task control unit 14 reduces the power consumption by the power shortage of 100 [Wh] by stopping the server P2 (S142).

FIG. 7 is a table showing what to do when the power shortage is 150 [Wh] in classification 3.
The task control unit 14 adds the power consumption of the servers in descending order of server priority until the power consumption corresponds to the power shortage. In the example of FIG. 5, the total power consumption of the three servers P2, P6, and P4 with lower server priorities is 160 [Wh], so the task control unit 14 targets these three servers to be stopped.

The task control unit 14 temporarily stores the task T7 on the server P6 to be stopped in the task storage unit 12 (S141). Similarly, the task control unit 14 may temporarily store the task T4 on the server P4 to be stopped in the task storage unit 12, but may also perform task replacement processing as described below.
(Step 1) This time, search for servers that are not subject to termination. Here, server P5, which has low power consumption and high power efficiency, was searched.
(Step 2) Among the tasks running on the server P5 (active tasks), a task T6 having a lower task priority than the task T4 to be replaced is found.
(Step 3) After temporarily storing task T6 on server P5 in task storage unit 12, task T4 is migrated (moved) to server P5. Due to this replacement process between tasks T6 and T4, the power consumption of server P5 will slightly increase by X from 50, but this slight increase is ignored in the following power consumption calculations as it is extremely smaller than the power consumption of the server itself. .

In addition, in this specification, it is assumed that up to two tasks can be operated at the same time per server, so by performing the process of saving task T6 in step 3 first, one free space for tasks on server P5 is prepared. did. On the other hand, if server P1 is selected with emphasis on processing performance in step 1, server P1 already has one empty task slot, so there is no need to save other tasks.
The task control unit 14 then stops the servers P2, P6, and P4 for which there are no active tasks, thereby reducing the amount of power used by the power shortage of 160 [Wh] (S142).

FIG. 8 is a table showing what to do when the power shortage is 200 [Wh] in classification 3.
The task control unit 14 adds the power consumption of the servers in descending order of server priority until the power consumption corresponds to the power shortage. In the example of FIG. 5, the total power consumption of the four servers P2, P6, P4, and P5 with lower server priorities is 210 [Wh], so the task control unit 14 targets these four servers to be stopped.

The task control unit 14 temporarily stores the following active tasks in the task storage unit 12 (S141).
・Task T7 on server P6 to be stopped
・Task T4 on server P4 to be stopped
・Tasks T5 and T6 on server P5 to be stopped
Thereafter, the task control unit 14 moves the task T5 having the highest task priority among the task groups stored in the task storage unit 12 to the server P1 that has one vacant task slot. As a result, the power consumption of server P5 slightly increases from 200 by Y.
The task control unit 14 then stops the servers P2, P6, P4, and P5 for which there are no active tasks, thereby reducing power consumption by the power shortage of 210 [Wh] (S142).

FIG. 9 is a table showing what to do when the available power recovers from the state shown in FIG. 8 and the power surplus state of classification 1 occurs.
The task control unit 14 restarts the stopped server P5 (S121). Then, the task control unit 14 reads tasks T4 and T6 with higher task priority among the suspended tasks T4, T6, and T7 from the task storage unit 12, and re-reserves the tasks to run on the server P5. (S122). Thereby, the number of operating tasks can be increased appropriately according to the amount of available power.

[effect]
The controller (task management device) 10 of the present invention includes:
a power generation amount collection unit 15 that calculates the usable power amount of the server group 21 from the power generation amount of the renewable power generation plant (power supply unit) 3;
a usage collection unit 13 that calculates the power usage of the server group 21;
a task control unit 14 that suspends the running task according to the amount of power shortage when the amount of available power is insufficient relative to the amount of power used, and stops the server that was running the suspended task; has.

Thereby, tasks can be appropriately allocated even in an unstable power supply source. For example, by using the regenerative power generation plant 3 as an unstable power supply source, it is possible to contribute to the improvement of global environmental problems by reducing carbon dioxide emissions. Further, even when using not only the regenerative power generation plant 3 but also a stable power supply source during normal operation, service provision can be continued in the event of a disaster such as a power outage.

The task control unit 14 calculates the priority of each task running in the server group 21 using predetermined input parameters, and also calculates the priority of each server in the server group 21 running the task from the priority of the task. Then, select the servers to be stopped in order of priority.

This allows high-priority tasks to continue operating even when there is a power shortage.

The task control unit 14 migrates the task on the server that will be stopped this time to the server that will not be stopped this time.

This makes it possible to shorten the service interruption period for tasks to be migrated.

The task control unit 14 saves, in the task storage unit 12, a snapshot at the time when the task on the server to be stopped this time is interrupted.

This saves the current state of the interrupted task, so there is no need to discard previous calculation results.

When the amount of available power recovers relative to the amount of power used, the task control unit 14 restarts the stopped server and relocates the tasks that had been saved in the task storage unit 12 to the restarted server. do.

This allows tasks with an appropriate load to be run according to the amount of available power, improving resource utilization efficiency.

1 Data center 3 Regenerative power plant (power supply section)
10 Controller (task management device)
11 Placement calculation unit 12 Task storage unit 13 Usage amount collection unit 14 Task control unit 15 Power generation amount collection unit 21 Server group 22 Battery 100 Task processing system

Claims (6)

a power generation amount collection unit that calculates the usable power amount of the server group based on the power amount obtained by subtracting the amount of power used to be stored in the battery from the power generation amount of the power supply unit, which is a regenerative power generation plant;
a usage collection unit that calculates the power usage of the server group;
Even if the stored power of the battery is used as a backup power source, if the available power amount is insufficient for the amount of power used, the currently running task is interrupted according to the amount of power shortage, and the interrupted task is and a task control unit that stops the server that was running the server and migrates the task on the server that is currently being stopped to a more power efficient server that is searched from a set of servers that will not be stopped this time. Task management device. The task control unit calculates the priority of each task running in the server group using predetermined input parameters, and calculates the priority of each server in the server group running the task from the priority of the task. The task management device according to claim 1, wherein servers are selected to be stopped in order of priority. 2. The task control unit saves, in the task storage unit, a snapshot taken at the time when a task on the power-efficient server that has a lower priority than the current task to be migrated is interrupted. task management device. The task control unit restarts the stopped server when the available power amount recovers with respect to the power usage amount, and transfers the task saved in the task storage unit to the restarted server. The task management device according to claim 3, wherein the task management device performs rearrangement. The task management device includes a power generation amount collection section, a usage amount collection section, and a task control section,
The power generation amount collecting unit calculates the usable power amount of the server group based on the power amount obtained by subtracting the power usage amount to be stored in the battery from the power generation amount of the power supply unit which is a regenerative power generation plant,
The usage collection unit calculates the power usage of the server group,
When the available power amount is insufficient for the amount of power used even if the stored power of the battery is used as a backup power source , the task control unit suspends the running task according to the amount of power shortage. At the same time, the server that was running the interrupted task is stopped, and the task on the server to be stopped this time is migrated to a more power efficient server searched from a set of servers that will not be stopped this time. Management method. A task management program for causing a computer to function as the task management device according to any one of claims 1 to 4.

Images